A method for optimizing catalyst preparation conditions based on machine learning and genetic algorithm

A method for optimizing catalyst preparation conditions based on machine learning and genetic algorithm | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A method for optimizing catalyst preparation conditions based on machine learning and genetic algorithm Junhui Ao, wei wang, ziyang Wang, Qianhong qiu, changming Du This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7555785/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Multiphase catalytic technology is an effective method for treating exhaust gases, but the traditional method for optimizing catalyst preparation conditions is inefficient and struggles to achieve multi-dimensional optimization. Aiming at the problems of low efficiency of experimental optimization and tend to fall into local optimum in the development of multiphase catalytic oxidation catalysts, we propose an intelligent optimization strategy integrating machine learning and genetic algorithm (GA). Taking Mn-Ca/γ-Al2O₃ as the research object, the effects of catalyst loading, loading ratio, calcination temperature and time on the degradation of xylene were systematically investigated, and a high-quality dataset comprising 225 sets of experimental data was established. Algorithms such as Random Forest (RF) and Gradient Boosting Tree (GBR) were used to establish a removal rate prediction model, which combined with SHAP values to resolve the key influencing factors, and we used GA to search for the optimal preparation parameters. The results showed that the RF model had the best prediction accuracy (R²=0.949, MAE=0.022 for the test set) and the characteristic importance showed that calcination temperature (43.9%) and load (21.96%) were the key control parameters. The best conditions obtained from GA optimization (5.4% loading, Ca: Mn=1:4, calcination temperature 423°C, time 3.75h) resulted in 93.7% xylene removal, which is a 2.5% enhancement over the conventional experimental optimization. This study provides a new paradigm for green and low-carbon development of catalysts. Catalyst optimization Machine learning Genetic algorithms Multiphase catalytic technology Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 1. Introduction End-of-pipe treatment of volatile organic compounds (VOCs) from industrial sources is a key component in realizing the synergies of pollution reduction and carbon reduction, and the main technologies include adsorption (e.g. activated carbon/zeolite rotor enrichment), absorption (chemical solvent scrubbing), combustion (RTO/RCO thermal oxidation), and biological treatment (biofilter degradation). Among them, multiphase catalytic oxidation technology is widely used because of its low temperature and high efficiency [ 1 ] . However, traditional catalyst R&D relies on an empirical trial-and-error approach, with long experimental cycles (usually 6–12 months) and high energy consumption (50-80kWh for single calcination), and tends to fall into local optimum [ 2 ][ 3 ] . Taking precious metal catalysts as an example, although they show excellent low-temperature activity, their widespread use is limited by the scarcity and high cost of noble metal resources, which makes it difficult to meet the needs of industrial-scale applications [ 4 ] . In recent years, transition metal composite catalysts have received attention for their low cost and high stability [ 5 ] , but the complex preparation parameters lead to an exponential expansion of the optimization space, and the traditional orthogonal experimental methods are no longer able to effectively address this challenge [ 6 ] . In this context, AI technology provides a new paradigm for catalyzing rational design. Random forest (RF) model significantly outperforms the conventional response surface method in predicting toluene oxidation activity of CuO-CeO₂ catalysts with R² value of 0.93 [ 7 ] . SHAP value analysis further showed that oxygen vacancy concentration contributed 38% to the regulation of the reaction pathway [ 8 ] . However, a single ML model is susceptible to dataset bias, and it is difficult to achieve global parameter optimization. Genetic algorithms (GA) show advantages in breaking through local optima due to their parallel search characteristics, such as reducing the CO oxidation energy barrier of Pt catalysts by 0.15 eV via surface reconstruction [ 9 ] .The GA-ANN framework proposed by Liu Siwei et al. improves the arrangement efficiency of dual catalysts by at least 95% [ 10 ] . Nevertheless, intelligent design of catalysts with asymmetric carriers still faces the challenge of insufficient model generalization [ 11 ] . In this study, a two-stage intelligent optimization framework of "feature analysis-global optimization" is proposed for Mn-Ca/γ-Al2O₃ catalysts. First, a high-dimensional dataset containing 225 sets of data was constructed by orthogonal and one-way experiments, covering key parameters such as loading, Ca/Mn ratio, calcination temperature and time. Random Forest (RF), Gradient Boosting Tree (GBR) and other algorithms are used to establish the removal rate prediction model, and combined with the SHAP value for feature importance resolution, breaking through the interpretability limitations of the traditional black-box models. Second, the RF model is used as the fitness function of the genetic algorithm to design a global search strategy for the multidimensional parameter space, which balances the exploration and exploitation contradictions through the elite retention and adaptive cross-variation mechanisms. Currently, the deep integration of intelligent algorithms and catalytic material design has made significant progress, such as the discovery of high-entropy oxides driven by active learning strategies to reduce the OER overpotential by 26% [ 12 ] , and DPA-2 large-atom model accurately analyzes the dynamic evolution of Fe 4 C active phase in Fe-FTS system [ 13 ] . This study not only verifies the universality of the ML-GA synergistic framework in the optimization of transition metal catalysts, but also provides a reusable technology paradigm for the intelligent transformation of industrial plants. In the future, this direction needs to further explore multi-objective optimization, dynamic working condition adaptive algorithms, and knowledge transfer mechanisms across material systems to achieve the whole chain of innovation from laboratory prediction to industrial scale-up. 2. Materials and methods 2.1 Experimental materials and equipment In the heterogeneous catalytic oxidation and material calcination experiments, o-xylene was used as the target pollutant, γ-Al 2 O₃ was used as the catalyst carrier, and reagents such as calcium nitrate tetrahydrate, silica particles, potassium sulfate, magnesium sulfate, sodium sulfate, and zinc sulfate heptahydrate were used to provide the doping elements or to assist the experiments. The experimental gases included nitrogen, o-xylene standard gas and air. The experimental equipment covers a packed spray tower for reactions, a peristaltic pump for transferring liquids, a ROSs generator for generating reactive oxygen molecules, a gas chromatograph for detecting the concentration of pollutants, as well as auxiliary equipment such as a hydrogen generator, a fully automated air generator, an electronic balance, a digital constant-temperature bath, an all-in-one intelligent muffle furnace, and an electric blast drying oven. 2.2 Catalyst preparation A spherical γ-Al₂O₃ (3–5mm in diameter) served as the carrier, with the total load of Ca and Mn elements set to 9% and the loading ratio fixed at 1:1. Firstly, 300 ml of salt solution was mixed with γ-Al2O₃ and stirred at 90°C in a water bath with heat until the liquid was evaporated. Next, the catalyst precursor was dried in an oven at 120°C for 6h, and finally, the catalyst was calcined in a muffle furnace at 4°C/min for 4h to produce the finished catalyst. 2.3 Machine learning algorithms Machine learning algorithms were used in this study to construct a predictive model for xylene removal rate, mainly including random forest (RF), gradient boosted tree (GBR), support vector regression (SVR) and ridge regression (Ridge) algorithms. In this study, these algorithms were used to model the nonlinear characteristics of environmental data, widely used in pollution prediction, real-time monitoring, and other scenarios. They learn the complex relationship between factors and xylene removal rates based on experimental data, and then predict the removal rates under different conditions Take the Random Forest algorithm as an example, it is an integrated learning algorithm based on the Bagging idea, where multiple training subsets are extracted from the original dataset with put-back, and some features are randomly selected for node splitting when constructing each decision tree. Each tree predicts the results independently, and the regression task takes the average output, so as to reduce the overfitting problem of a single tree and improve the generalization ability of the model [ 14 ][ 15 ] . In this study, the random forest model learns and analyzes several factors that affect the xylene removal rate, such as catalyst loading, loading ratio, calcination temperature and time, and then makes a prediction of the removal rate. This study uses a combination of learning curves and grid search (GridSearchCV) to determine the model hyperparameters. Learning curves show the effect of parameter changes on the model to obtain optimal values; Grid search enumerates different combinations of hyperparameters and searches for values that optimize the model's performance [ 15 ] . Evaluate model performance using metrics such as R², MSE, RMSE, and MAE, which measure the difference between the model's predicted value and the true value from different perspectives, to comprehensively assess the model's predictive performance. By comparison, Random Forest (RF) and Gradient Boosting Tree (GBR) are found to be the most effective and have similar training and testing performance, with better generalization ability [ 15 ] . Among them, Random Forest (RF) has a strong nonlinear fitting ability, is robust to noise and outliers, and performs superiorly in the regression task [ 17 ] ; Support Vector Regression (SVR) has weak nonlinear fitting ability and performs poorly when dealing with complex data distributions [ 18 ] ; Ridge regression (Ridge), as a linear model, is mainly suitable for dealing with data with linear relationships, and lacks the ability to capture nonlinear relationships, but is more stable [ 19 ] . Taken together, the Random Forest (RF) model performed best in the prediction of xylene removal rate in this study, and thus was used in the subsequent genetic algorithm optimization process as a fitness model to search for the optimal catalyst preparation parameter combinations [ 20 ] . 3. Results & Discussion 3.1 Results and analysis of multiple catalytic oxidative degradation of xylene The impact of volatile organic compounds (VOCs) emissions from industrial sources on greenhouse gas emissions has become an important environmental challenge as global concerns about climate change and environmental sustainability rise. VOCs generate ozone and fine particulate matter through photochemical reactions in the atmosphere, exacerbating air pollution, and their oxidation process indirectly contributes to carbon dioxide emissions, exacerbating global warming. As a typical pollutant of VOCs, xylene is emitted in the chemical, petroleum refining and printing industries, posing a serious threat to the environment and human health. In this chapter, the degradation effect of Mn-Ca/γ-Al2O₃ catalysts was evaluated with xylene as a representative, and the effects of catalyst loading, reactive oxygen species (ROS) injection, active component loading and ratio, and catalyst calcination temperature and time on the degradation efficiency were investigated, which provided data support for the subsequent optimization of the process conditions for catalyst preparation. 3.1.1 Effect of catalytic packing loading on the degradation of xylene Catalyst loading has a significant impact on the efficiency of xylene degradation in terms of cost control and maximization of the removal rate. The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L·min, and catalyst loading of 6% for a continuous reaction of 100 min, and adjusting the Mn-Ca/γ-Al2O₃ catalytic filler loading to 50 g, 70 g, 90 g, 120 g, and 150 g, respectively. As shown in Fig. 2 , the catalyst loading is positively correlated with the removal efficiency, and the increase in removal rate diminishes when the loading amount exceeds 90 g, with only a slight increase to 94.51% at 150 g. During the reaction, the catalyst removal rate for each loading amount decreased due to adsorption of reactants and formation of by-products on the catalyst surface, leading to inactivation of active sites. However, the higher loading catalysts were able to maintain a higher removal rate for a longer period of time, for example, at 100 minutes, the removal rate of the 150g loading catalyst was still 42.81% compared to 4.27% for the 50g loading. In practical industrial applications, a reasonable regeneration strategy should be formulated by considering the catalyst dosage, deactivation rate and economic benefits. 3.1.2 Effect of ROSs injection on the degradation of xylene ROSs are used as oxidants in multiphase catalytic oxidation systems, and their injection amount directly affects pollutant degradation. The experiments were carried out with a xylene concentration of 500 ppm, an airflow of 3 L/min, a catalyst loading of 6%, a catalytic packing of 90 g, and a continuous reaction for 100 min, and the ROSs injection amounts were adjusted to be 0mg/L·min, 20mg/L·min, 40mg/L·min, 60mg/L·min, 80mg/L·min, and 100mg/L·min, respectively. Figure 3 shows that the injection amount of ROSs had a significant effect on the removal efficiency of xylene, which showed a tendency of increasing and then decreasing, and the optimal removal rate was achieved when the injection amount was increased to 80 mg/L·min. Increasing the injection amount of ROSs can improve the contact probability with xylene and promote the direct oxidative degradation of xylene in the gas phase, which is also conducive to the generation of more ·OH under the effect of compound catalysis and the participation of degradation. However, when the concentration of ROSs exceeds 80mg/L·min, the removal rate decreases due to the saturation of catalyst active sites and competition with xylene for active sites. 3.1.3 Effect of catalyst loading on the degradation of xylene The amount of loading determines the number of active sites of the catalyst, and the appropriate amount of loading can improve the reaction rate and removal rate. The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L·min, catalytic packing capacity of 90 g, and continuous reaction for 100 min, and the total loading was adjusted to 3%, 6%, 9%, 12%, and 15%, respectively. As shown in Fig. 4 , the pollutant removal increased and then decreased with the increase of loading within 60 minutes, and the catalyst performance was optimal at 6% loading, with a removal efficiency of 93.79%. Below 6%, insufficient loading results in a paucity of active sites and a decrease in the removal rate. Above 6%, excessive loading of active components will cause aggregation of active sites, reduce the specific surface area and pore volume of the catalyst, reduce the effective active sites, and affect the removal efficiency [ 23 ] . After 60 min, the catalysts with loading greater than 9% were more effective, which was due to the fact that in the initial stage, the active sites aggregated, and some of the sites were not effectively exposed. At the later stage of the reaction the reactants diffuse into the interior and more active sites are utilized. 3.1.4 Effect of catalyst ratio on xylene degradation Doping ratio is a key parameter in catalyst design, and precise control can improve overall performance. .The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L/min, and catalytic packing volume of 90 g. The continuous reaction was carried out for 100 min, and the doping ratios of Ca and Mn were adjusted to be 1:4, 1:2, 1:1, 2:1, and 4:1, respectively. As shown in Fig. 5 , the catalyst removal efficiency was optimal at the 1:1 doping ratio, when the basic and acidic sites were in equilibrium, providing an ideal chemical environment for xylene degradation. Increasing the ratio of Ca to Mn to 4:1 results in better catalyst stability, reduced replacement frequency, and lower operating costs. 3.1.5 Effect of catalyst calcination temperature on the degradation of xylene Suitable calcination temperature can retain the catalyst activity and stability, and improve the application effect. The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L/min, catalyst loading of 9%, catalytic packing capacity of 90 g. The reaction was carried out continuously for 100 min, and the calcinations temperatures of the catalysts were adjusted to be 200°C, 350°C, 500°C, 650°C, and 800°C, respectively. Figure 6 shows that The catalytic efficiency of the catalyst is excellent at 350°C and 500°C, with excellent stability and slow decay of the removal rate at 350°C, which is attributed to the stable lattice structure formed at this temperature [ 24 ] . The initial removal rate of the catalyst calcined at 200°C was high, but decreased sharply with reaction time because the active sites were not sufficiently activated and the precursors were not completely decomposed [ 24 ] . The significant decrease in catalyst performance at 800°C calcination is due to the fusion and aggregation of catalyst particles at high temperatures, the shrinkage of specific surface area, the decrease in the number of active centers, and the change in surface chemistry [ 26 ] . 3.1.6 Effect of catalyst calcination time on the degradation of xylene Calcination time has a significant effect on the physical and chemical properties of catalysts. The experiments were carried out with a controlled airflow of 3 L/min, a toluene concentration of 500 ppm, a catalytic packing capacity of 90 g, an injection of ROSs of 60 mg/L·min, a catalyst loading of 9%, a calcination temperature of 500°C, a continuous reaction for 100 min, and adjusted catalyst calcinations of 2 h, 4 h, 6 h, 8 h, and 10 h, respectively. As shown in Fig. 7 , the catalyst removal rate decreased significantly with the prolongation of calcination time, which was attributed to the fact that too long calcination time promoted sintering between catalyst particles, reduced the specific surface area, decreased the number of active sites and accessibility, and may also result in the inability of metal oxide catalysts to form active crystalline phases with high catalytic activity. The best overall performance of the catalyst was achieved with a 2-hour calcination, with a high initial removal rate, and a high removal rate of 47.31% was maintained at 100 minutes. The rapid decrease in stability of catalysts calcined at 6 h may be related to excessive migration of active components, irreversible structural changes, or disruption of the pore structure. 3.2 Dataset preparation The AI algorithm predicts better in the application domain, and the distribution of sample data in the hyperplane of the preparation condition optimization model should be as uniform as possible. Orthogonal experiments are statistical methods based on orthogonal tables that ensure a balanced distribution of levels for each factor and reduce data bias. In this study, we consider the application domain principle, conduct orthogonal experiments (Table 1 ), and combine the data from the 3.1 Influence Factors Exploration Experiment, which together serve as the training data for the prediction models. Table 1 Orthogonal experiments Serial number Loading capacity (%) Loading proportion (%) Calcination temperature (℃) Calcination time (h) Removal rate 1 0.15 1.00 500 6 0.714771858 2 0.12 1.00 500 6 0.783615487 3 0.09 1.00 500 6 0.803955515 4 0.06 1.00 500 6 0.837187017 5 0.03 1.00 500 6 0.807267791 6 0.09 0.50 500 6 0.715653175 7 0.09 0.25 500 6 0.732295872 8 0.09 1.00 500 6 0.803955515 9 0.09 2.00 500 6 0.769199646 10 0.09 4.00 500 6 0.773502691 11 0.09 0.04 200 6 0.736385157 12 0.09 1.00 350 6 0.921058271 13 0.09 2.00 500 6 0.803955515 14 0.09 3.00 650 6 0.712870677 15 0.09 4.00 800 6 0.54225449 16 0.09 5.00 500 2 0.885194471 17 0.09 6.00 500 4 0.797098277 18 0.09 7.00 500 6 0.803955515 19 0.09 8.00 500 8 0.80355081 20 0.09 1.00 500 10 0.797515956 21 0.03 0.50 200 2 0.81200777 22 0.03 0.25 500 8 0.929751063 23 0.03 1.00 800 4 0.490185984 24 0.03 2.00 350 10 0.895381353 25 0.03 4.00 650 6 0.804445475 26 0.06 0.50 800 8 0.659931963 27 0.06 0.25 350 4 0.93687005 28 0.06 1.00 650 10 0.670637678 29 0.06 2.00 200 6 0.773482714 30 0.06 4.00 500 2 0.894877218 31 0.09 0.50 650 4 0.827521063 32 0.09 0.25 200 10 0.85197786 33 0.09 1.00 500 6 0.870778903 34 0.09 2.00 800 2 0.741208363 35 0.09 4.00 350 8 0.837646381 36 0.12 0.50 500 10 0.817050651 37 0.12 0.25 800 6 0.713464597 38 0.12 1.00 350 2 0.783167277 39 0.12 2.00 650 8 0.733363852 40 0.12 4.00 200 4 0.38181198 41 0.15 0.50 350 6 0.807348531 42 0.15 0.25 650 2 0.742400919 43 0.15 1.00 200 8 0.607873669 44 0.15 2.00 500 4 0.694553905 45 0.15 4.00 800 10 0.337721462 3.3 Machine Learning Predicts Results Construction of a predictive model for xylene removal rate using Random Forest (RF), Gradient Boosted Tree (GBR), Support Vector Regression (SVR) and Ridge Regression (Ridge) algorithms. During model training, the choice of hyperparameters is critical to model performance. In this study, the method of combining Learning Curve and GridSearchCV is used to determine the hyperparameters used in model training, and the specific hyperparameter settings are shown in Table 2 . Table 2 Hyperparameters of the removal rate prediction machine learning model Machine Learning hyperparameter GBR num_leaves : 6 max_depth : 5 boosting_type:'gbdt' objective : 'regression' random_state : 0 Machine Learning hyperparameter RF n_estimators: 100 min_samples_split = 2 min_samples_leaf = 1 random_state: 0 SVM Kernel : "auto" C : 0.1 epsilon : 0.1 Ridge alpHa: 0.1 max_iter: 10 tol: 0 Learning curves can show how the model's performance changes as parameters are varied to obtain optimal values. Since different hyperparameters interact with each other, the lattice search searches for the hyperparameter values that optimize the model's performance by enumerating different combinations of hyperparameters. Table 3 Prediction results of removal rate of machine learning model Model Dataset R 2 MSE MAE RF Train 0.985 0.0002 0.009 Test 0.949 0.0008 0.022 GBR Train 0.974 0.0004 0.014 Test 0.941 0.0009 0.024 SVR Train 0.223 0.0120 0.080 Test 0.175 0.0128 0.086 Ridge Train 0.687 0.0048 0.062 Test 0.637 0.0056 0.066 Evaluate model performance using metrics such as R², MSE, RMSE and MAE. The results of the predictive model are shown in Table 3 . R² is used to measure the goodness of fit of the regression model, and takes a value ranging from 0 to 1. The closer it is to 1, the better the model fits the data, i.e., the better the independent variable explains the dependent variable [ 27 ] ; MSE reflects the mean squared deviation of the prediction error, and the closer its value is to 0, the smaller the difference between the model's predicted value and the true value is, and the better the model is fitted [ 28 ] ; RMSE is the square root of MSE, also used to measure the degree of dispersion of the prediction error, the smaller the value of RMSE, the higher the prediction accuracy of the model. MAE quantifies the average absolute magnitude of the prediction error, and the closer its value is to 0, the smaller the deviation of the model's predicted value from the true value is [ 15 ] . The formula for calculating these indicators is as follows: $$\:{R}^{2}=1-\frac{{\frac{1}{n}{\sum\:}_{i=1}^{n}\left({Y}_{i}-{y}_{i}\right)}^{2}}{{\frac{1}{n}{\sum\:}_{i=1}^{n}\left({Y}_{i}-{\stackrel{-}{Y}}_{i}\right)}^{2}}\:Equation\:\left(1\right)$$ $$\:\text{M}\text{S}\text{E}={\frac{1}{n}{\sum\:}_{i=1}^{n}\left({Y}_{i}-{y}_{i}\right)}^{2\:}Equation\:\left(2\right)$$ $$\:\text{R}\text{M}\text{S}\text{E}=\sqrt{{\frac{1}{n}{\sum\:}_{i=1}^{n}\left({Y}_{i}-{y}_{i}\right)}^{2}\:}Equation\:\left(3\right)$$ $$\:\text{M}\text{A}\text{E}=\frac{1}{n}{\sum\:}_{i=1}^{n}\left|{Y}_{i}-{y}_{i}\right|\:Equation\:\left(4\right)$$ where, Y—— the true value of the data; y—— the predicted value of the model; \(\:\stackrel{-}{Y}\) —— the average value; n—— the number of samples. The training data were expanded to 225 sample sizes using data enhancement methods, 90% of which were used as the training set for model training and optimization, during which the loss function was minimized by adjusting the model parameters; The remaining 10% is used as a test set to evaluate the model's performance on unseen data, and to judge the model's generalization ability and prediction accuracy by calculating various performance metrics. 3.4 Genetic Algorithm (GA) Optimization Genetic Algorithm (GA) with the objective of maximizing pollutant removal rate, and the samples obtained from orthogonal experiments were used as the initial population. Orthogonal experiment is a statistical method based on orthogonal tables, which ensures a balanced distribution of levels for each factor, reduces data bias, and enables the initial population to evenly cover the optimization space of calcination temperature, calcination time, load, and load ratio. The removal rate prediction function obtained from the random forest model is used as a fitness function as a way to assess the merit of individual solutions. The fitness function plays a key role in genetic algorithms by providing a criterion for comparing the strengths and weaknesses of individuals, and the algorithm selects the best individuals to pass on to the next generation based on the fitness value. As shown in Fig. 8 , the core operations of genetic algorithms include crossover, mutation and selection. The crossover operation uses a single-point crossover, in which two individuals of the parent generation are randomly selected during each generation update and their gene segments are exchanged at the selected crossover position, thus generating a new individual, and this approach helps to explore a wider solution space and to inherit the good characteristics of the parent generation. The mutation operation randomly mutates the candidate offspring with a mutation rate of 20%, simulating the phenomenon of genetic mutation in biological evolution, introducing new genetic information into the population, and preventing the algorithm from falling into a local optimum solution [ 21 ] . Select operations to use the elite retention strategy [ 22 ][ 30 ] , where 50% of each of the parent and child generations are preferred as the next generation, and the most adapted parent individual is directly retained to ensure the continuation of the current optimal solution and accelerate the algorithm's convergence to the global optimal solution. The optimization conditions were set as loading, loading ratio, calcination temperature and calcination time, the number of populations per generation was set as 9, and the number of iterative updates was 100. During the operation of the algorithm, these parameters are continuously adjusted to search for catalyst preparation formulations that maximize the pollutant removal rate. In each iteration, the strengths and weaknesses of each individual are evaluated by the fitness function, and the best individuals are selected for crossover and mutation operations to generate a new population that gradually approaches the global optimal solution. In this study, the genetic algorithm was used to maximize pollutant removal rate as the optimization objective, and load, load ratio, calcination temperature, and calcination time as the optimization conditions, with the number of populations in each generation set to 9 and the number of iterative updates set to 100. The final search yielded optimal reaction conditions of 0.054 loading, 0.25 loading ratio, calcination temperature of 423.438°C, calcination time of 3.745 h, and a predicted xylene removal rate of 0.9366. As verified by multiphase catalytic oxidation experiments, the catalyst removal rate obtained by genetic algorithm optimization was 0.9373, which was significantly higher than that of the catalyst obtained by conventional experimental optimization of 0.9120. This indicates that the genetic algorithm is able to search effectively in multi-dimensional space to achieve better parameter combinations, avoid falling into local optimums and significantly improve the overall performance of the catalysts, as well as reduce the carbon consumption in the process of catalyst development, as compared with the traditional experimental optimization methods. 3.5 Character analysis In this part, the Random Forest (RF) model combined with the SHAP value was used to analyze the importance of each feature in depth in order to investigate the effect of catalyst preparation conditions on its performance. The random forest (RF) model measures feature importance by the degree of reduction in impurity when nodes split, and the SHAP value quantifies the marginal contribution of each feature to the model output across different feature combinations [ 29 ] . According to the analysis results of Fig. 9 and Fig. 10 , the calcination temperature has the most critical effect on the catalyst performance, contributing 43.9% to the removal rate, and the SHAP value indicates that the catalyst performance is optimal at a calcination temperature of 350–500°C, which is consistent with the experimental results of 3.1, where inappropriate temperatures will reduce the catalyst activity. The loading was the next most important, contributing 21.96%, and the SHAP value showed a negative correlation with the catalyst performance, which, combined with the results of the 3.1 experiment, showed that loading over 9% resulted in a significant decrease in performance. The loading ratio and calcination time also have some influence, although there is no linear relationship with the removal rate, but in general, a higher Ca and Mn loading ratio and shorter calcination time within a certain range are favorable to improve the catalyst performance. The characteristic significance analysis of SHAP values clarified the dominant roles of calcination temperature and loading, and combined with the experimental results, revealed the mechanism of variable effects, providing a scientific basis for optimizing the performance of catalysts. 3.6 Experimental validation In the experiment of 3.1, the optimal preparation conditions obtained were: catalyst loading of 6%, loading ratio of 1:1, calcination temperature of 350 degrees Celsius, calcination time of 2h. While the preparation conditions obtained by genetic algorithm (GA) were: loading of 0.054, loading ratio of 0.25, calcination temperature of 423.438°C, calcination time of 3.745 h, and predicted xylene removal rate of 0.9366. Experimental validation of multiphase catalytic oxidation for evaluating the effectiveness of genetic algorithms (GA) and random forest models. The experimental conditions were kept the same as in the previous section, the xylene concentration was 500 ppm, the ROSs concentration was 60 mg/L, and the catalyst used was 100 g. A total of three parallel experiments were carried out, and each time the reaction time was 30 minutes, and the average removal rate was taken for each 10-minute period. The final experimental results showed that the removal of xylene by the catalyst prepared with the optimal parameters obtained experimentally was 0.9120, while the removal of xylene by the catalyst prepared with the optimal conditions obtained through genetic algorithm (GA) search was 0.9373. It can be seen that the optimal parameters obtained experimentally are prone to fall into local optimization due to the interaction of various conditions, while genetic algorithms (GA) are more likely to obtain global optimization through selection, mutation, and crossover, and they can save carbon consumption in the process of catalyst development. 4. Conclusion The end-of-pipe treatment of volatile organic compounds (VOCs) from industrial sources is extremely critical to the synergy of pollution reduction and carbon reduction. Although multiphase catalytic oxidation technology is widely used, the traditional catalyst research and development is faced with the problems of long cycle time, high energy consumption, and easy to be caught in the local optimum. In this study, an optimization strategy incorporating machine learning and genetic algorithm (GA) is proposed for Mn-Ca/γ-Al2O₃. Construction of a high-quality dataset of 225 data sets by orthogonal and one-way experiments, and we established a xylene removal rate prediction model using random forest (RF) and gradient boosting tree (GBR) algorithms, among which the RF model had the best prediction accuracy (R²=0.949, MAE = 0.022 for the test set). The RF model was used as the fitness function and a global search was performed using GA to obtain the optimal preparation conditions (5.4% loading, Ca:Mn = 1:4, calcination temperature 423°C, and time 3.75h), which resulted in a 93.7% xylene removal rate, an improvement of 2.5% compared to the traditional experimental optimization. Combined RF model and SHAP value analysis revealed that calcination temperature (43.9%) and load (21.96%) were the key control parameters. Experimental results show that GA can avoid falling into local optimum, enhance overall catalyst performance, and reduce carbon consumption. This study provides a new paradigm for the green and low-carbon development of catalysts in the field of heterogeneous catalysis, and mechanisms such as multi-objective optimization can be further explored in the future to promote industrial chain-wide innovation. Declarations Declaration of Competing Interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Changming Du reports financial support was provided by National Natural Science Foundation of China. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Funding The work is supported by the National Natural Science Foundation of China [61871409]. Author Contribution A.J. wrote the main manuscript text.W.W. was responsible for visualization and verification.W.Z.Y. was in charge of part of the image drawing.Q.Q.H. proposed the concept.D.C.M. was responsible for the review. References Du, C. M., Lu, S. Y., & Ding, J. M. (2023). Heterogeneous catalytic oxidation treatment of organic waste gas and engineering applications. Beijing, China: Chemical Industry Press. Guo, Y. L., Wen, M. C., Li, G. Y., & An, T. C. (2021). Recent advances in VOC elimination by catalytic oxidation technology onto various nanoparticles catalysts: A critical review. Applied Catalysis B: Environmental, 281(2), 119447. Konsolakis, M. (2015). Recent advances in catalyst technology for volatile organic compound (VOC) removal. Catalysis Today, 258, 13–23. https://doi.org/10.1016/j.cattod.2015.03.041 Wang, H., Zhang, L., Li, X., et al. (2022). Non-noble metal catalysts for catalytic oxidation of volatile organic compounds: A review. Applied Catalysis B: Environmental, 310, 121336 .https://doi.org/10.1016/j.apcatb.2022.121336 Zeng, Y. Q., Haw, K. G., Wang, Z. G., Wang, Y. A., Zhang, S. L., & Hongmanorom, P. (2021). Double redox process to synthesize CuO-CeO₂ catalysts with strong Cu-Ce interaction for efficient toluene oxidation. Journal of Hazardous Materials, 404, 124088. Chen, Y., Li, R., Suo, H., & Zhang, Z. (2021). Evaluation of a data-driven machine learning approach for identifying potential candidates for environmental catalysts: From database development to prediction. ACS ES&T Engineering, 1(8), 1246–1257. Shih, S., Lee, W.J., Lin, L.F., Huang, C.H. (2012). Prediction of catalytic activity of Cu-Ce/γ-Al2O3catalysts by artificial neural networks. Chemical Engineering Journal, 181–182, 130–136. https://doi.org/10.1016/j.cej.2011.11.040 Lundberg, S. M., & Lee, S. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30, 4768–4777. Liu, Y., Dai, H., Deng, J., Xie, S., Yang, H., Jiang, Y. (2015). Surface reconstruction of Pt nanoparticles for enhanced CO oxidation: A combined experimental and DFT study. ACS Catalysis, 5(2), 909–919. https://doi.org/10.1021/cs501668k Liu, S., Wen, C., Liu, Q., et al. (2024). Theoretical optimization of bed packing arrangement in cascade Dual-Catalyst system with side reactions. Chemical Engineering Science, 299, 120500. https://doi.org/10.1016/j.ces.2024.120500 Wang, Z. L., Zhang, H. K., Ren, J. H., & Li, X. (2021). Predicting adsorption ability of adsorbents at arbitrary sites for pollutants using deep transfer learning. npj Computational Materials, 7(1), 19. Jablonka, K. M., Ongari, D., Moosavi, S. M., & Smit, B. (2020).Big-data science in porous materials: Materials genomics and machine learning.Chemical Reviews, 120(16), 8066–8129. https://doi.org/10.1021/acs.chemrev.0c00004 Jiao, F., Li, J., Pan, X., Xiao, J., Li, H. (2016). Selective conversion of syngas to light olefins. Science, 351(6277), 1065–1068. https://doi.org/10.1126/science.aaf1835 Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. Li, X., Zhang, Y., Cheng, J., Liu, Y. (2019). Machine learning-assisted design of porous materials for VOC adsorption. Microporous and Mesoporous Materials, 278, 379–386. https://doi.org/10.1016/j.micromeso.2019.01.010 Hastie, T., Tibshirani, R., & Friedman, J. (2009). The elements of statistical learning: Data mining, inference, and prediction (2nd ed.). Springer. Dong, Y. T. (2023). Comparison of Random Forest and Traditional Classical Method in Regression and Classification Problems. Statistics and Applications, 12(2), 255–260. https://doi.org/10.12677/SA.2023.122026 Drucker, H., Burges, C. J. C., Kaufman, L., Smola, A., & Vapnik, V. (1997). Support vector regression machines. Advances in Neural Information Processing Systems, 9, 155–161. Montgomery, D. C., Peck, E. A., & Vining, G. G. (2021). Introduction to linear regression analysis (6th ed.). Wiley. Shin, Y., Kim, Z., Yu, J., & Park, S. (2019). Development of NOx reduction system utilizing artificial neural network and genetic algorithm. Journal of Cleaner Production, 232, 1418–1429. Goldberg, D. E. (1989). Genetic Algorithms in Search, Optimization and Machine Learning. Addison-Wesley. Mitchell, M. (1996). An Introduction to Genetic Algorithms. MIT Press. Rezaei, E., Soltan, J., Chen, N. (2013). Catalytic oxidation of toluene by ozone over alumina supported manganese oxides: Effect of catalyst loading. Applied Catalysis B: Environmental, 136–137, 239–247. https://doi.org/10.1016/j.apcatb.2013.01.060 Wang, Y., Arandiyan, H., Scott, J., Amal, R., Dai, H., Deng, J., Li, J. (2017). Effect of calcination temperature on the activity and structure of MnOx/γ-Al2O3 catalysts for formaldehyde oxidation. Catalysis Today, 281, 467–472. https://doi.org/10.1016/j.cattod.2016.05.044 Zhang, L., He, C., Li, Y., Cheng, Y. (2020). Recent advances in catalytic ozonation for water treatment. Chemical Engineering Journal, 397, 125421. https://doi.org/10.1016/j.cej.2020.125421 Li, F., Wang, S., Li, T., Tian, Y., Wang, M., & Cai, W. (2023). Effect of Calcination Temperature on the Performance of SiO 2 @Co@CeO 2 Catalyst in CO 2 Reforming With Ethanol. Catalysis Letters, 153(12), 3712–3723. https://doi.org/10.1007/s10562-023-04282-6 Draper, N. R. Smith, H. Applied Regression Analysis, 3rd Edition[M]. New York: Springer. 2014. James, G., Witten, D., Hastie, T., & Tibshirani, R. (2021).An Introduction to Statistical Learning: With Applications in R (2nd ed.).Springer. https://doi.org/10.1007/978-1-0716-1418-1 Lundberg, S. M., Erion, G., Chen, H., DeGrave, A., Prutkin, J. M., Nair, B., … Lee, S. I. (2020).From local explanations to global understanding with explainable AI for trees.Nature Machine Intelligence, 2(1), 56–67. https://doi.org/10.1038/s42256-019-0138-9 Raccuglia, P., Elbert, K. C., Adler, P. D., Falk, C., Wenny, M. B., Mollo, A., … Brinson,L. C. (2016).Machine-learning-assisted materials discovery using failed experiments.Nature,533(7601), 73–76.https://doi.org/10.1038/nature17439 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7555785","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":533957686,"identity":"7499a02c-a74f-49b5-96ac-fa9402f505a4","order_by":0,"name":"Junhui Ao","email":"","orcid":"","institution":"SUN YAT-SEN UNIVERSITY","correspondingAuthor":false,"prefix":"","firstName":"Junhui","middleName":"","lastName":"Ao","suffix":""},{"id":533957687,"identity":"ca19420c-0ec4-4f86-acef-ee4efd5521f7","order_by":1,"name":"wei wang","email":"","orcid":"","institution":"Taizhou Huanfa Environmental Technology Co., Ltd","correspondingAuthor":false,"prefix":"","firstName":"wei","middleName":"","lastName":"wang","suffix":""},{"id":533957688,"identity":"bd1e6699-1db7-4db2-9b6f-b04ae96c8eb3","order_by":2,"name":"ziyang Wang","email":"","orcid":"","institution":"SUN YAT-SEN UNIVERSITY","correspondingAuthor":false,"prefix":"","firstName":"ziyang","middleName":"","lastName":"Wang","suffix":""},{"id":533957693,"identity":"eb5d90d7-0577-45e1-bbf3-652cada6aaae","order_by":3,"name":"Qianhong qiu","email":"","orcid":"","institution":"SUN YAT-SEN UNIVERSITY","correspondingAuthor":false,"prefix":"","firstName":"Qianhong","middleName":"","lastName":"qiu","suffix":""},{"id":533957694,"identity":"deb99cd9-924d-4134-b1c7-88f4c26a489c","order_by":4,"name":"changming Du","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA5UlEQVRIiWNgGAWjYDCCAwxsDAwVDAxs7HChBGK0nAFqYSZJC2MbkEG0Fr4byc8e3Zy3TZ6PmYH5M8+fwwz87DkGDD934NYieSPN3Dh3223DNmYGNmnetsMMkj1vDBh7z+DWYnAjh00aqIURpIWZt+EwSMSAGexUvFrm3LZvgznMnjgtDbcTgVoYpHnYgLZIENAieeaZmXTOsdvJbUBlknPb0nkkzjwrONiLRwvf8eRn0jk1t23ntzcf/vDmj7Ucf3vyxgc/8WhBAowNTDwMDDwg5gGiNIA1/SBa6SgYBaNgFIwkAAA59EwIrv9ERgAAAABJRU5ErkJggg==","orcid":"","institution":"SUN YAT-SEN UNIVERSITY","correspondingAuthor":true,"prefix":"","firstName":"changming","middleName":"","lastName":"Du","suffix":""}],"badges":[],"createdAt":"2025-09-07 11:08:07","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7555785/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7555785/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":94479711,"identity":"af9500b2-f725-4d72-8882-2c15a7c1af71","added_by":"auto","created_at":"2025-10-27 16:07:33","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":362955,"visible":true,"origin":"","legend":"","description":"","filename":"MLandGAforMnCaAlOCatalystOptimization..docx","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/e221219d9059ccb1fba69df1.docx"},{"id":94479864,"identity":"194bdfa7-e5a5-4fbb-8c61-8d420b120bde","added_by":"auto","created_at":"2025-10-27 16:08:14","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6491,"visible":true,"origin":"","legend":"","description":"","filename":"3f03c64554444921a81f2aa24b3178bf.json","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/0ba460f81d18aa1a6a061c16.json"},{"id":94480180,"identity":"8aa699cf-da5c-4ea5-8353-96c98c0b71d9","added_by":"auto","created_at":"2025-10-27 16:10:06","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":113006,"visible":true,"origin":"","legend":"","description":"","filename":"3f03c64554444921a81f2aa24b3178bf1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/2d6c399a8a8958e72cb96aae.xml"},{"id":94480215,"identity":"d6ca6790-8435-4222-baa2-c5ace8d68828","added_by":"auto","created_at":"2025-10-27 16:10:07","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":68225,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/a6d9142f68e584be417de675.png"},{"id":94479745,"identity":"9dade61c-50f6-4944-ab81-51f73b2f5d77","added_by":"auto","created_at":"2025-10-27 16:07:46","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19703,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/175d69efec777da2274b28f6.png"},{"id":94480091,"identity":"f9ffdbbb-9c16-48ef-8d21-fed55463b7fa","added_by":"auto","created_at":"2025-10-27 16:09:32","extension":"emf","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":117300,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.emf","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/deab0a0c5dfb3f937ed4f693.emf"},{"id":94479887,"identity":"1a474155-6ff8-4364-8bd9-569ef1d4ffbb","added_by":"auto","created_at":"2025-10-27 16:08:38","extension":"emf","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":66040,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.emf","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/215f1a8212ed08709935315c.emf"},{"id":94480086,"identity":"93c4332a-9dca-482a-ad71-4191dcaf88d8","added_by":"auto","created_at":"2025-10-27 16:09:27","extension":"emf","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59372,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.emf","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/8adf8f26073266905b86a2c2.emf"},{"id":94480068,"identity":"ae3008b6-903f-4a39-8389-7bae13af0dd1","added_by":"auto","created_at":"2025-10-27 16:09:19","extension":"emf","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59420,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.emf","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/05e146ce4751f4e260842661.emf"},{"id":94479660,"identity":"b85b2a06-dad6-4381-80f8-97f0aeb53350","added_by":"auto","created_at":"2025-10-27 16:07:06","extension":"emf","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59532,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage6.emf","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/44474e34e6392effcdbeb6ba.emf"},{"id":94479709,"identity":"fc2af2eb-9580-41c8-9a58-243edb87116a","added_by":"auto","created_at":"2025-10-27 16:07:32","extension":"emf","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59284,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.emf","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/d5331579f5a0daf2164af7b6.emf"},{"id":94480087,"identity":"dad01394-85e0-4e4e-b1d9-d84f3dbc12d9","added_by":"auto","created_at":"2025-10-27 16:09:28","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":29018,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/6b2691a3efa39710a3395ae3.png"},{"id":94479808,"identity":"ef472e2c-6422-4a16-bcf3-f0a8ea29e1a5","added_by":"auto","created_at":"2025-10-27 16:07:52","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":114644,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/856e016e170556ddfe4ce29e.png"},{"id":94480057,"identity":"f508b7a4-e26e-487c-ac17-bfc38c238fca","added_by":"auto","created_at":"2025-10-27 16:09:12","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15322,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/75dad2922da412a72d2c5167.png"},{"id":94480168,"identity":"fbbb7114-a36b-40ef-9d56-7680946aac29","added_by":"auto","created_at":"2025-10-27 16:10:01","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14279,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/ec199262b87575c45c8b556e.png"},{"id":94479846,"identity":"f3463175-51e9-44f6-876f-a6ecc14c44db","added_by":"auto","created_at":"2025-10-27 16:08:03","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40265,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/050d7e0873051467b07ef281.png"},{"id":94480088,"identity":"1cd751e5-b8e2-4dc0-8dba-7a22e3713352","added_by":"auto","created_at":"2025-10-27 16:09:29","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":34848,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/c7718ae4add0974a5750bc33.png"},{"id":94479870,"identity":"b88675a9-4e86-4932-be4e-c8949918ecbf","added_by":"auto","created_at":"2025-10-27 16:08:19","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":32089,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/b1c4cc9154e28e387f3b3d43.png"},{"id":94480217,"identity":"1a4789c4-451a-4f3e-ac8d-28a44b5b7b61","added_by":"auto","created_at":"2025-10-27 16:10:07","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":30493,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/68efe93c721b0630b1e49742.png"},{"id":94479712,"identity":"9f88e191-d0ab-4a28-a48a-f20acf9efc04","added_by":"auto","created_at":"2025-10-27 16:07:33","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":35775,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/effe6a2c766bedda9c2108a0.png"},{"id":94489720,"identity":"f5eb6d58-4b0f-4d86-be75-8fa02a98f11c","added_by":"auto","created_at":"2025-10-27 17:05:46","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":31075,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/fcfa06f2a958e766608225ff.png"},{"id":94479716,"identity":"6715f14c-7d9e-492b-b6e2-61c6adb0b66d","added_by":"auto","created_at":"2025-10-27 16:07:33","extension":"png","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15636,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/bb5c4385fadf53bdf1080f28.png"},{"id":94479863,"identity":"4531bbc6-cd15-4763-82c1-e1c165778ee8","added_by":"auto","created_at":"2025-10-27 16:08:14","extension":"png","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":21807,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/1e19a0827e05a6a2894eafb7.png"},{"id":94479662,"identity":"81933b44-fd5f-4c2c-963a-fc4efea1b276","added_by":"auto","created_at":"2025-10-27 16:07:09","extension":"xml","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":111317,"visible":true,"origin":"","legend":"","description":"","filename":"3f03c64554444921a81f2aa24b3178bf1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/040f218faeea781d4a59eda0.xml"},{"id":94479717,"identity":"ff490002-0b74-48c7-bae7-8ee51d0d79ea","added_by":"auto","created_at":"2025-10-27 16:07:34","extension":"html","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":118000,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/810f425c2c04a424e6ec5b09.html"},{"id":94479594,"identity":"7d582f68-170c-4d9c-be95-b3b54ec62786","added_by":"auto","created_at":"2025-10-27 16:06:53","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":68225,"visible":true,"origin":"","legend":"\u003cp\u003eCatalyst Preparation Process\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/6ccc975a6b643d0819ad0881.png"},{"id":94479733,"identity":"73782131-c9f5-4354-8c5f-ae4d746c4c76","added_by":"auto","created_at":"2025-10-27 16:07:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":554143,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of catalyst loading on xylene removal rate\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/ce79b3119504d01ab8bc9735.png"},{"id":94479806,"identity":"c5dc04a6-cb61-4c57-91fa-bcd31fa9b29b","added_by":"auto","created_at":"2025-10-27 16:07:51","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":591194,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of ROSs on xylene removal\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/c62afe9dd06bbcf0a15452bb.png"},{"id":94479703,"identity":"00dc3af8-4742-49ad-80e8-63c71c622dc3","added_by":"auto","created_at":"2025-10-27 16:07:30","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":543859,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of catalyst loading on xylene removal\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/9ae160067f10c24a502ffe4c.png"},{"id":94479892,"identity":"2dbb6336-f360-45e8-87fc-d3698699a62d","added_by":"auto","created_at":"2025-10-27 16:08:44","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":542915,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of active component loading ratio on xylene removal rate\u003c/p\u003e","description":"","filename":"image5.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/543cbebc3d53a37f53e9e901.png"},{"id":94479737,"identity":"c3eb3ea6-b6b1-481f-8ecd-38ac7e7d6a92","added_by":"auto","created_at":"2025-10-27 16:07:39","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":579702,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of catalyst calcination temperature on xylene removal rate\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/efed04a23643ef42d45bdf95.png"},{"id":94480056,"identity":"62a931c9-c94e-479a-bfbe-e2e701625144","added_by":"auto","created_at":"2025-10-27 16:09:11","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":534756,"visible":true,"origin":"","legend":"\u003cp\u003eEffect of catalyst calcination time on xylene removal rate\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/126219fb73403c8c69b59a74.png"},{"id":94479974,"identity":"13719c82-8c10-4a53-a144-02c9c46eeae9","added_by":"auto","created_at":"2025-10-27 16:08:59","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":29018,"visible":true,"origin":"","legend":"\u003cp\u003eGenetic algorithm (GA) iteration process\u003c/p\u003e","description":"","filename":"image8.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/8df4c1e0bcb03e55f7c1fc05.png"},{"id":94480077,"identity":"20848f19-73f2-4d26-ae1c-0a8071e54d54","added_by":"auto","created_at":"2025-10-27 16:09:24","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":114644,"visible":true,"origin":"","legend":"\u003cp\u003eSHAP analysis results\u003c/p\u003e","description":"","filename":"image9.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/07d429e8b2c7146024714c76.png"},{"id":94479521,"identity":"f5e6a30a-8840-45ea-a5aa-049bd5b6ac71","added_by":"auto","created_at":"2025-10-27 16:06:07","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":19703,"visible":true,"origin":"","legend":"\u003cp\u003eRandom forest feature importance\u003c/p\u003e","description":"","filename":"image10.png","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/d676fc23c4b92280f15753b7.png"},{"id":94987294,"identity":"c31187d7-1462-4be6-a183-9d130453c842","added_by":"auto","created_at":"2025-11-03 07:01:41","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":4670319,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7555785/v1/8dacfdaf-7286-4b9c-86a5-3a73fd7ee5d1.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A method for optimizing catalyst preparation conditions based on machine learning and genetic algorithm","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eEnd-of-pipe treatment of volatile organic compounds (VOCs) from industrial sources is a key component in realizing the synergies of pollution reduction and carbon reduction, and the main technologies include adsorption (e.g. activated carbon/zeolite rotor enrichment), absorption (chemical solvent scrubbing), combustion (RTO/RCO thermal oxidation), and biological treatment (biofilter degradation).\u003c/p\u003e\u003cp\u003eAmong them, multiphase catalytic oxidation technology is widely used because of its low temperature and high efficiency\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. However, traditional catalyst R\u0026amp;D relies on an empirical trial-and-error approach, with long experimental cycles (usually 6\u0026ndash;12 months) and high energy consumption (50-80kWh for single calcination), and tends to fall into local optimum\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e][\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. Taking precious metal catalysts as an example, although they show excellent low-temperature activity, their widespread use is limited by the scarcity and high cost of noble metal resources, which makes it difficult to meet the needs of industrial-scale applications\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e. In recent years, transition metal composite catalysts have received attention for their low cost and high stability\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e, but the complex preparation parameters lead to an exponential expansion of the optimization space, and the traditional orthogonal experimental methods are no longer able to effectively address this challenge\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn this context, AI technology provides a new paradigm for catalyzing rational design. Random forest (RF) model significantly outperforms the conventional response surface method in predicting toluene oxidation activity of CuO-CeO₂ catalysts with R\u0026sup2; value of 0.93 \u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. SHAP value analysis further showed that oxygen vacancy concentration contributed 38% to the regulation of the reaction pathway\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e. However, a single ML model is susceptible to dataset bias, and it is difficult to achieve global parameter optimization. Genetic algorithms (GA) show advantages in breaking through local optima due to their parallel search characteristics, such as reducing the CO oxidation energy barrier of Pt catalysts by 0.15 eV via surface reconstruction\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e.The GA-ANN framework proposed by Liu Siwei et al. improves the arrangement efficiency of dual catalysts by at least 95%\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. Nevertheless, intelligent design of catalysts with asymmetric carriers still faces the challenge of insufficient model generalization\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003cp\u003eIn this study, a two-stage intelligent optimization framework of \"feature analysis-global optimization\" is proposed for Mn-Ca/γ-Al2O₃ catalysts. First, a high-dimensional dataset containing 225 sets of data was constructed by orthogonal and one-way experiments, covering key parameters such as loading, Ca/Mn ratio, calcination temperature and time. Random Forest (RF), Gradient Boosting Tree (GBR) and other algorithms are used to establish the removal rate prediction model, and combined with the SHAP value for feature importance resolution, breaking through the interpretability limitations of the traditional black-box models. Second, the RF model is used as the fitness function of the genetic algorithm to design a global search strategy for the multidimensional parameter space, which balances the exploration and exploitation contradictions through the elite retention and adaptive cross-variation mechanisms.\u003c/p\u003e\u003cp\u003eCurrently, the deep integration of intelligent algorithms and catalytic material design has made significant progress, such as the discovery of high-entropy oxides driven by active learning strategies to reduce the OER overpotential by 26%\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e, and DPA-2 large-atom model accurately analyzes the dynamic evolution of Fe\u003csub\u003e4\u003c/sub\u003eC active phase in Fe-FTS system\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e. This study not only verifies the universality of the ML-GA synergistic framework in the optimization of transition metal catalysts, but also provides a reusable technology paradigm for the intelligent transformation of industrial plants. In the future, this direction needs to further explore multi-objective optimization, dynamic working condition adaptive algorithms, and knowledge transfer mechanisms across material systems to achieve the whole chain of innovation from laboratory prediction to industrial scale-up.\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1 Experimental materials and equipment\u003c/h2\u003e\u003cp\u003eIn the heterogeneous catalytic oxidation and material calcination experiments, o-xylene was used as the target pollutant, γ-Al\u003csub\u003e2\u003c/sub\u003eO₃ was used as the catalyst carrier, and reagents such as calcium nitrate tetrahydrate, silica particles, potassium sulfate, magnesium sulfate, sodium sulfate, and zinc sulfate heptahydrate were used to provide the doping elements or to assist the experiments. The experimental gases included nitrogen, o-xylene standard gas and air.\u003c/p\u003e\u003cp\u003eThe experimental equipment covers a packed spray tower for reactions, a peristaltic pump for transferring liquids, a ROSs generator for generating reactive oxygen molecules, a gas chromatograph for detecting the concentration of pollutants, as well as auxiliary equipment such as a hydrogen generator, a fully automated air generator, an electronic balance, a digital constant-temperature bath, an all-in-one intelligent muffle furnace, and an electric blast drying oven.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2 Catalyst preparation\u003c/h2\u003e\u003cp\u003eA spherical γ-Al₂O₃ (3\u0026ndash;5mm in diameter) served as the carrier, with the total load of Ca and Mn elements set to 9% and the loading ratio fixed at 1:1. Firstly, 300 ml of salt solution was mixed with γ-Al2O₃ and stirred at 90\u0026deg;C in a water bath with heat until the liquid was evaporated. Next, the catalyst precursor was dried in an oven at 120\u0026deg;C for 6h, and finally, the catalyst was calcined in a muffle furnace at 4\u0026deg;C/min for 4h to produce the finished catalyst.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3 Machine learning algorithms\u003c/h2\u003e\u003cp\u003eMachine learning algorithms were used in this study to construct a predictive model for xylene removal rate, mainly including random forest (RF), gradient boosted tree (GBR), support vector regression (SVR) and ridge regression (Ridge) algorithms. In this study, these algorithms were used to model the nonlinear characteristics of environmental data, widely used in pollution prediction, real-time monitoring, and other scenarios. They learn the complex relationship between factors and xylene removal rates based on experimental data, and then predict the removal rates under different conditions\u003c/p\u003e\u003cp\u003eTake the Random Forest algorithm as an example, it is an integrated learning algorithm based on the Bagging idea, where multiple training subsets are extracted from the original dataset with put-back, and some features are randomly selected for node splitting when constructing each decision tree. Each tree predicts the results independently, and the regression task takes the average output, so as to reduce the overfitting problem of a single tree and improve the generalization ability of the model\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e][\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. In this study, the random forest model learns and analyzes several factors that affect the xylene removal rate, such as catalyst loading, loading ratio, calcination temperature and time, and then makes a prediction of the removal rate.\u003c/p\u003e\u003cp\u003eThis study uses a combination of learning curves and grid search (GridSearchCV) to determine the model hyperparameters. Learning curves show the effect of parameter changes on the model to obtain optimal values; Grid search enumerates different combinations of hyperparameters and searches for values that optimize the model's performance\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. Evaluate model performance using metrics such as R\u0026sup2;, MSE, RMSE, and MAE, which measure the difference between the model's predicted value and the true value from different perspectives, to comprehensively assess the model's predictive performance.\u003c/p\u003e\u003cp\u003eBy comparison, Random Forest (RF) and Gradient Boosting Tree (GBR) are found to be the most effective and have similar training and testing performance, with better generalization ability\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. Among them, Random Forest (RF) has a strong nonlinear fitting ability, is robust to noise and outliers, and performs superiorly in the regression task\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/sup\u003e; Support Vector Regression (SVR) has weak nonlinear fitting ability and performs poorly when dealing with complex data distributions \u003csup\u003e[\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e; Ridge regression (Ridge), as a linear model, is mainly suitable for dealing with data with linear relationships, and lacks the ability to capture nonlinear relationships, but is more stable\u003csup\u003e[\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e. Taken together, the Random Forest (RF) model performed best in the prediction of xylene removal rate in this study, and thus was used in the subsequent genetic algorithm optimization process as a fitness model to search for the optimal catalyst preparation parameter combinations\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Results \u0026 Discussion","content":"\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\n\u003ch2\u003e3.1 Results and analysis of multiple catalytic oxidative degradation of xylene\u003c/h2\u003e\n\u003cp\u003eThe impact of volatile organic compounds (VOCs) emissions from industrial sources on greenhouse gas emissions has become an important environmental challenge as global concerns about climate change and environmental sustainability rise. VOCs generate ozone and fine particulate matter through photochemical reactions in the atmosphere, exacerbating air pollution, and their oxidation process indirectly contributes to carbon dioxide emissions, exacerbating global warming. As a typical pollutant of VOCs, xylene is emitted in the chemical, petroleum refining and printing industries, posing a serious threat to the environment and human health. In this chapter, the degradation effect of Mn-Ca/\u0026gamma;-Al2O₃ catalysts was evaluated with xylene as a representative, and the effects of catalyst loading, reactive oxygen species (ROS) injection, active component loading and ratio, and catalyst calcination temperature and time on the degradation efficiency were investigated, which provided data support for the subsequent optimization of the process conditions for catalyst preparation.\u003c/p\u003e\n\u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\n\u003ch2\u003e3.1.1 Effect of catalytic packing loading on the degradation of xylene\u003c/h2\u003e\n\u003cp\u003eCatalyst loading has a significant impact on the efficiency of xylene degradation in terms of cost control and maximization of the removal rate. The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L\u0026middot;min, and catalyst loading of 6% for a continuous reaction of 100 min, and adjusting the Mn-Ca/\u0026gamma;-Al2O₃ catalytic filler loading to 50 g, 70 g, 90 g, 120 g, and 150 g, respectively. As shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e, the catalyst loading is positively correlated with the removal efficiency, and the increase in removal rate diminishes when the loading amount exceeds 90 g, with only a slight increase to 94.51% at 150 g. During the reaction, the catalyst removal rate for each loading amount decreased due to adsorption of reactants and formation of by-products on the catalyst surface, leading to inactivation of active sites. However, the higher loading catalysts were able to maintain a higher removal rate for a longer period of time, for example, at 100 minutes, the removal rate of the 150g loading catalyst was still 42.81% compared to 4.27% for the 50g loading. In practical industrial applications, a reasonable regeneration strategy should be formulated by considering the catalyst dosage, deactivation rate and economic benefits.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec9\" class=\"Section3\"\u003e\n\u003ch2\u003e3.1.2 Effect of ROSs injection on the degradation of xylene\u003c/h2\u003e\n\u003cp\u003eROSs are used as oxidants in multiphase catalytic oxidation systems, and their injection amount directly affects pollutant degradation. The experiments were carried out with a xylene concentration of 500 ppm, an airflow of 3 L/min, a catalyst loading of 6%, a catalytic packing of 90 g, and a continuous reaction for 100 min, and the ROSs injection amounts were adjusted to be 0mg/L\u0026middot;min, 20mg/L\u0026middot;min, 40mg/L\u0026middot;min, 60mg/L\u0026middot;min, 80mg/L\u0026middot;min, and 100mg/L\u0026middot;min, respectively. Figure\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e shows that the injection amount of ROSs had a significant effect on the removal efficiency of xylene, which showed a tendency of increasing and then decreasing, and the optimal removal rate was achieved when the injection amount was increased to 80 mg/L\u0026middot;min. Increasing the injection amount of ROSs can improve the contact probability with xylene and promote the direct oxidative degradation of xylene in the gas phase, which is also conducive to the generation of more \u0026middot;OH under the effect of compound catalysis and the participation of degradation. However, when the concentration of ROSs exceeds 80mg/L\u0026middot;min, the removal rate decreases due to the saturation of catalyst active sites and competition with xylene for active sites.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\n\u003ch2\u003e3.1.3 Effect of catalyst loading on the degradation of xylene\u003c/h2\u003e\n\u003cp\u003eThe amount of loading determines the number of active sites of the catalyst, and the appropriate amount of loading can improve the reaction rate and removal rate. The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L\u0026middot;min, catalytic packing capacity of 90 g, and continuous reaction for 100 min, and the total loading was adjusted to 3%, 6%, 9%, 12%, and 15%, respectively. As shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e4\u003c/span\u003e, the pollutant removal increased and then decreased with the increase of loading within 60 minutes, and the catalyst performance was optimal at 6% loading, with a removal efficiency of 93.79%. Below 6%, insufficient loading results in a paucity of active sites and a decrease in the removal rate. Above 6%, excessive loading of active components will cause aggregation of active sites, reduce the specific surface area and pore volume of the catalyst, reduce the effective active sites, and affect the removal efficiency\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/sup\u003e. After 60 min, the catalysts with loading greater than 9% were more effective, which was due to the fact that in the initial stage, the active sites aggregated, and some of the sites were not effectively exposed. At the later stage of the reaction the reactants diffuse into the interior and more active sites are utilized.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\n\u003ch2\u003e3.1.4 Effect of catalyst ratio on xylene degradation\u003c/h2\u003e\n\u003cp\u003eDoping ratio is a key parameter in catalyst design, and precise control can improve overall performance. .The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L/min, and catalytic packing volume of 90 g. The continuous reaction was carried out for 100 min, and the doping ratios of Ca and Mn were adjusted to be 1:4, 1:2, 1:1, 2:1, and 4:1, respectively. As shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e5\u003c/span\u003e, the catalyst removal efficiency was optimal at the 1:1 doping ratio, when the basic and acidic sites were in equilibrium, providing an ideal chemical environment for xylene degradation. Increasing the ratio of Ca to Mn to 4:1 results in better catalyst stability, reduced replacement frequency, and lower operating costs.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section3\"\u003e\n\u003ch2\u003e3.1.5 Effect of catalyst calcination temperature on the degradation of xylene\u003c/h2\u003e\n\u003cp\u003eSuitable calcination temperature can retain the catalyst activity and stability, and improve the application effect. The experiments were carried out by controlling the xylene concentration of 500 ppm, air flow rate of 3 L/min, ROSs injection rate of 60 mg/L/min, catalyst loading of 9%, catalytic packing capacity of 90 g. The reaction was carried out continuously for 100 min, and the calcinations temperatures of the catalysts were adjusted to be 200\u0026deg;C, 350\u0026deg;C, 500\u0026deg;C, 650\u0026deg;C, and 800\u0026deg;C, respectively. Figure\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e shows that The catalytic efficiency of the catalyst is excellent at 350\u0026deg;C and 500\u0026deg;C, with excellent stability and slow decay of the removal rate at 350\u0026deg;C, which is attributed to the stable lattice structure formed at this temperature\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e. The initial removal rate of the catalyst calcined at 200\u0026deg;C was high, but decreased sharply with reaction time because the active sites were not sufficiently activated and the precursors were not completely decomposed\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/sup\u003e. The significant decrease in catalyst performance at 800\u0026deg;C calcination is due to the fusion and aggregation of catalyst particles at high temperatures, the shrinkage of specific surface area, the decrease in the number of active centers, and the change in surface chemistry\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section3\"\u003e\n\u003ch2\u003e3.1.6 Effect of catalyst calcination time on the degradation of xylene\u003c/h2\u003e\n\u003cp\u003eCalcination time has a significant effect on the physical and chemical properties of catalysts. The experiments were carried out with a controlled airflow of 3 L/min, a toluene concentration of 500 ppm, a catalytic packing capacity of 90 g, an injection of ROSs of 60 mg/L\u0026middot;min, a catalyst loading of 9%, a calcination temperature of 500\u0026deg;C, a continuous reaction for 100 min, and adjusted catalyst calcinations of 2 h, 4 h, 6 h, 8 h, and 10 h, respectively. As shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e, the catalyst removal rate decreased significantly with the prolongation of calcination time, which was attributed to the fact that too long calcination time promoted sintering between catalyst particles, reduced the specific surface area, decreased the number of active sites and accessibility, and may also result in the inability of metal oxide catalysts to form active crystalline phases with high catalytic activity. The best overall performance of the catalyst was achieved with a 2-hour calcination, with a high initial removal rate, and a high removal rate of 47.31% was maintained at 100 minutes. The rapid decrease in stability of catalysts calcined at 6 h may be related to excessive migration of active components, irreversible structural changes, or disruption of the pore structure.\u003c/p\u003e\n\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n\u003ch2\u003e3.2 Dataset preparation\u003c/h2\u003e\n\u003cp\u003eThe AI algorithm predicts better in the application domain, and the distribution of sample data in the hyperplane of the preparation condition optimization model should be as uniform as possible. Orthogonal experiments are statistical methods based on orthogonal tables that ensure a balanced distribution of levels for each factor and reduce data bias. In this study, we consider the application domain principle, conduct orthogonal experiments (Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e), and combine the data from the 3.1 Influence Factors Exploration Experiment, which together serve as the training data for the prediction models.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab1\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eOrthogonal experiments\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003eSerial number\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eLoading capacity (%)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eLoading proportion (%)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eCalcination temperature (℃)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eCalcination time (h)\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eRemoval rate\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e1\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.714771858\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.783615487\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e3\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.803955515\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.06\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.837187017\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e5\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.807267791\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.50\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.715653175\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e7\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.732295872\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.803955515\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e9\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.769199646\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.773502691\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e11\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.04\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e200\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.736385157\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e350\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.921058271\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e13\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.803955515\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e14\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e3.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e650\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.712870677\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e800\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.54225449\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e16\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e5.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.885194471\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e17\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.797098277\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e18\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e7.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.803955515\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e19\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.80355081\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e20\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.797515956\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e21\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.50\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e200\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.81200777\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e22\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.929751063\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e23\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e800\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.490185984\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e24\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e350\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.895381353\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.03\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e650\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.804445475\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e26\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.06\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.50\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e800\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.659931963\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e27\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.06\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e350\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.93687005\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e28\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.06\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e650\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.670637678\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e29\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.06\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e200\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.773482714\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e30\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.06\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.894877218\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e31\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.50\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e650\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.827521063\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e32\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e200\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.85197786\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e33\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.870778903\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e34\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e800\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.741208363\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e35\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.09\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e350\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.837646381\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e36\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.50\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.817050651\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e37\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e800\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.713464597\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e38\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e350\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.783167277\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e39\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e650\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.733363852\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e40\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.12\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e200\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.38181198\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e41\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.50\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e350\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e6\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.807348531\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e42\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.25\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e650\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.742400919\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e43\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e1.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e200\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e8\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.607873669\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e44\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e2.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e500\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.694553905\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e45\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd colspan=\"2\" align=\"left\"\u003e\n\u003cp\u003e0.15\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e4.00\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e800\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e10\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.337721462\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\n\u003ch2\u003e3.3 Machine Learning Predicts Results\u003c/h2\u003e\n\u003cp\u003eConstruction of a predictive model for xylene removal rate using Random Forest (RF), Gradient Boosted Tree (GBR), Support Vector Regression (SVR) and Ridge Regression (Ridge) algorithms. During model training, the choice of hyperparameters is critical to model performance. In this study, the method of combining Learning Curve and GridSearchCV is used to determine the hyperparameters used in model training, and the specific hyperparameter settings are shown in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab2\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003eHyperparameters of the removal rate prediction machine learning model\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMachine Learning\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003ehyperparameter\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eGBR\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003enum_leaves : 6\u003c/p\u003e\n\u003cp\u003emax_depth : 5\u003c/p\u003e\n\u003cp\u003eboosting_type:'gbdt'\u003c/p\u003e\n\u003cp\u003eobjective : 'regression'\u003c/p\u003e\n\u003cp\u003erandom_state : 0\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eMachine Learning\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ehyperparameter\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRF\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003en_estimators: 100\u003c/p\u003e\n\u003cp\u003emin_samples_split\u0026thinsp;=\u0026thinsp;2\u003c/p\u003e\n\u003cp\u003emin_samples_leaf\u0026thinsp;=\u0026thinsp;1\u003c/p\u003e\n\u003cp\u003erandom_state: 0\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eSVM\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eKernel : \"auto\"\u003c/p\u003e\n\u003cp\u003eC : 0.1\u003c/p\u003e\n\u003cp\u003eepsilon : 0.1\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eRidge\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003ealpHa: 0.1\u003c/p\u003e\n\u003cp\u003emax_iter: 10\u003c/p\u003e\n\u003cp\u003etol: 0\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eLearning curves can show how the model's performance changes as parameters are varied to obtain optimal values. Since different hyperparameters interact with each other, the lattice search searches for the hyperparameter values that optimize the model's performance by enumerating different combinations of hyperparameters.\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n\u003ctable id=\"Tab3\" border=\"1\"\u003e\u003ccaption\u003e\n\u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n\u003cdiv class=\"CaptionContent\"\u003e\n\u003cp\u003ePrediction results of removal rate of machine learning model\u003c/p\u003e\n\u003c/div\u003e\n\u003c/caption\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eModel\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eDataset\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eR\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMSE\u003c/p\u003e\n\u003c/th\u003e\n\u003cth align=\"left\"\u003e\n\u003cp\u003eMAE\u003c/p\u003e\n\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003eRF\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTrain\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.985\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0002\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.009\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTest\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.949\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0008\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.022\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003eGBR\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTrain\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.974\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0004\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.014\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTest\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.941\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0009\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.024\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003eSVR\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTrain\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.223\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0120\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.080\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTest\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.175\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0128\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.086\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd rowspan=\"2\" align=\"left\"\u003e\n\u003cp\u003eRidge\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTrain\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.687\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0048\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.062\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003eTest\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.637\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.0056\u003c/p\u003e\n\u003c/td\u003e\n\u003ctd align=\"left\"\u003e\n\u003cp\u003e0.066\u003c/p\u003e\n\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eEvaluate model performance using metrics such as R\u0026sup2;, MSE, RMSE and MAE. The results of the predictive model are shown in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. R\u0026sup2; is used to measure the goodness of fit of the regression model, and takes a value ranging from 0 to 1. The closer it is to 1, the better the model fits the data, i.e., the better the independent variable explains the dependent variable\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e; MSE reflects the mean squared deviation of the prediction error, and the closer its value is to 0, the smaller the difference between the model's predicted value and the true value is, and the better the model is fitted\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e; RMSE is the square root of MSE, also used to measure the degree of dispersion of the prediction error, the smaller the value of RMSE, the higher the prediction accuracy of the model. MAE quantifies the average absolute magnitude of the prediction error, and the closer its value is to 0, the smaller the deviation of the model's predicted value from the true value is \u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. The formula for calculating these indicators is as follows:\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n\u003cdiv id=\"FileID_Equa\" class=\"mathdisplay\"\u003e$$\\:{R}^{2}=1-\\frac{{\\frac{1}{n}{\\sum\\:}_{i=1}^{n}\\left({Y}_{i}-{y}_{i}\\right)}^{2}}{{\\frac{1}{n}{\\sum\\:}_{i=1}^{n}\\left({Y}_{i}-{\\stackrel{-}{Y}}_{i}\\right)}^{2}}\\:Equation\\:\\left(1\\right)$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Equb\" class=\"Equation\"\u003e\n\u003cdiv id=\"FileID_Equb\" class=\"mathdisplay\"\u003e$$\\:\\text{M}\\text{S}\\text{E}={\\frac{1}{n}{\\sum\\:}_{i=1}^{n}\\left({Y}_{i}-{y}_{i}\\right)}^{2\\:}Equation\\:\\left(2\\right)$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Equc\" class=\"Equation\"\u003e\n\u003cdiv id=\"FileID_Equc\" class=\"mathdisplay\"\u003e$$\\:\\text{R}\\text{M}\\text{S}\\text{E}=\\sqrt{{\\frac{1}{n}{\\sum\\:}_{i=1}^{n}\\left({Y}_{i}-{y}_{i}\\right)}^{2}\\:}Equation\\:\\left(3\\right)$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Equd\" class=\"Equation\"\u003e\n\u003cdiv id=\"FileID_Equd\" class=\"mathdisplay\"\u003e$$\\:\\text{M}\\text{A}\\text{E}=\\frac{1}{n}{\\sum\\:}_{i=1}^{n}\\left|{Y}_{i}-{y}_{i}\\right|\\:Equation\\:\\left(4\\right)$$\u003c/div\u003e\n\u003c/div\u003e\n\u003cp\u003ewhere, Y\u0026mdash;\u0026mdash; the true value of the data;\u003c/p\u003e\n\u003cp\u003ey\u0026mdash;\u0026mdash; the predicted value of the model;\u003c/p\u003e\n\u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\stackrel{-}{Y}\\)\u003c/span\u003e\u003c/span\u003e\u0026mdash;\u0026mdash; the average value;\u003c/p\u003e\n\u003cp\u003en\u0026mdash;\u0026mdash; the number of samples.\u003c/p\u003e\n\u003cp\u003eThe training data were expanded to 225 sample sizes using data enhancement methods, 90% of which were used as the training set for model training and optimization, during which the loss function was minimized by adjusting the model parameters; The remaining 10% is used as a test set to evaluate the model's performance on unseen data, and to judge the model's generalization ability and prediction accuracy by calculating various performance metrics.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\n\u003ch2\u003e3.4 Genetic Algorithm (GA) Optimization\u003c/h2\u003e\n\u003cp\u003eGenetic Algorithm (GA) with the objective of maximizing pollutant removal rate, and the samples obtained from orthogonal experiments were used as the initial population. Orthogonal experiment is a statistical method based on orthogonal tables, which ensures a balanced distribution of levels for each factor, reduces data bias, and enables the initial population to evenly cover the optimization space of calcination temperature, calcination time, load, and load ratio.\u003c/p\u003e\n\u003cp\u003eThe removal rate prediction function obtained from the random forest model is used as a fitness function as a way to assess the merit of individual solutions. The fitness function plays a key role in genetic algorithms by providing a criterion for comparing the strengths and weaknesses of individuals, and the algorithm selects the best individuals to pass on to the next generation based on the fitness value.\u003c/p\u003e\n\u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e8\u003c/span\u003e, the core operations of genetic algorithms include crossover, mutation and selection. The crossover operation uses a single-point crossover, in which two individuals of the parent generation are randomly selected during each generation update and their gene segments are exchanged at the selected crossover position, thus generating a new individual, and this approach helps to explore a wider solution space and to inherit the good characteristics of the parent generation. The mutation operation randomly mutates the candidate offspring with a mutation rate of 20%, simulating the phenomenon of genetic mutation in biological evolution, introducing new genetic information into the population, and preventing the algorithm from falling into a local optimum solution\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e. Select operations to use the elite retention strategy\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e22\u003c/span\u003e][\u003cspan class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e, where 50% of each of the parent and child generations are preferred as the next generation, and the most adapted parent individual is directly retained to ensure the continuation of the current optimal solution and accelerate the algorithm's convergence to the global optimal solution.\u003c/p\u003e\n\u003cp\u003eThe optimization conditions were set as loading, loading ratio, calcination temperature and calcination time, the number of populations per generation was set as 9, and the number of iterative updates was 100. During the operation of the algorithm, these parameters are continuously adjusted to search for catalyst preparation formulations that maximize the pollutant removal rate. In each iteration, the strengths and weaknesses of each individual are evaluated by the fitness function, and the best individuals are selected for crossover and mutation operations to generate a new population that gradually approaches the global optimal solution.\u003c/p\u003e\n\u003cp\u003eIn this study, the genetic algorithm was used to maximize pollutant removal rate as the optimization objective, and load, load ratio, calcination temperature, and calcination time as the optimization conditions, with the number of populations in each generation set to 9 and the number of iterative updates set to 100. The final search yielded optimal reaction conditions of 0.054 loading, 0.25 loading ratio, calcination temperature of 423.438\u0026deg;C, calcination time of 3.745 h, and a predicted xylene removal rate of 0.9366. As verified by multiphase catalytic oxidation experiments, the catalyst removal rate obtained by genetic algorithm optimization was 0.9373, which was significantly higher than that of the catalyst obtained by conventional experimental optimization of 0.9120. This indicates that the genetic algorithm is able to search effectively in multi-dimensional space to achieve better parameter combinations, avoid falling into local optimums and significantly improve the overall performance of the catalysts, as well as reduce the carbon consumption in the process of catalyst development, as compared with the traditional experimental optimization methods.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\n\u003ch2\u003e3.5 Character analysis\u003c/h2\u003e\n\u003cp\u003eIn this part, the Random Forest (RF) model combined with the SHAP value was used to analyze the importance of each feature in depth in order to investigate the effect of catalyst preparation conditions on its performance. The random forest (RF) model measures feature importance by the degree of reduction in impurity when nodes split, and the SHAP value quantifies the marginal contribution of each feature to the model output across different feature combinations\u003csup\u003e[\u003cspan class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e. According to the analysis results of Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e9\u003c/span\u003e and Fig.\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e10\u003c/span\u003e, the calcination temperature has the most critical effect on the catalyst performance, contributing 43.9% to the removal rate, and the SHAP value indicates that the catalyst performance is optimal at a calcination temperature of 350\u0026ndash;500\u0026deg;C, which is consistent with the experimental results of 3.1, where inappropriate temperatures will reduce the catalyst activity. The loading was the next most important, contributing 21.96%, and the SHAP value showed a negative correlation with the catalyst performance, which, combined with the results of the 3.1 experiment, showed that loading over 9% resulted in a significant decrease in performance. The loading ratio and calcination time also have some influence, although there is no linear relationship with the removal rate, but in general, a higher Ca and Mn loading ratio and shorter calcination time within a certain range are favorable to improve the catalyst performance. The characteristic significance analysis of SHAP values clarified the dominant roles of calcination temperature and loading, and combined with the experimental results, revealed the mechanism of variable effects, providing a scientific basis for optimizing the performance of catalysts.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\n\u003ch2\u003e3.6 Experimental validation\u003c/h2\u003e\n\u003cp\u003eIn the experiment of 3.1, the optimal preparation conditions obtained were: catalyst loading of 6%, loading ratio of 1:1, calcination temperature of 350 degrees Celsius, calcination time of 2h. While the preparation conditions obtained by genetic algorithm (GA) were: loading of 0.054, loading ratio of 0.25, calcination temperature of 423.438\u0026deg;C, calcination time of 3.745 h, and predicted xylene removal rate of 0.9366. Experimental validation of multiphase catalytic oxidation for evaluating the effectiveness of genetic algorithms (GA) and random forest models. The experimental conditions were kept the same as in the previous section, the xylene concentration was 500 ppm, the ROSs concentration was 60 mg/L, and the catalyst used was 100 g. A total of three parallel experiments were carried out, and each time the reaction time was 30 minutes, and the average removal rate was taken for each 10-minute period. The final experimental results showed that the removal of xylene by the catalyst prepared with the optimal parameters obtained experimentally was 0.9120, while the removal of xylene by the catalyst prepared with the optimal conditions obtained through genetic algorithm (GA) search was 0.9373. It can be seen that the optimal parameters obtained experimentally are prone to fall into local optimization due to the interaction of various conditions, while genetic algorithms (GA) are more likely to obtain global optimization through selection, mutation, and crossover, and they can save carbon consumption in the process of catalyst development.\u003c/p\u003e\n\u003c/div\u003e"},{"header":"4. Conclusion","content":"\u003cp\u003eThe end-of-pipe treatment of volatile organic compounds (VOCs) from industrial sources is extremely critical to the synergy of pollution reduction and carbon reduction. Although multiphase catalytic oxidation technology is widely used, the traditional catalyst research and development is faced with the problems of long cycle time, high energy consumption, and easy to be caught in the local optimum. In this study, an optimization strategy incorporating machine learning and genetic algorithm (GA) is proposed for Mn-Ca/γ-Al2O₃. Construction of a high-quality dataset of 225 data sets by orthogonal and one-way experiments, and we established a xylene removal rate prediction model using random forest (RF) and gradient boosting tree (GBR) algorithms, among which the RF model had the best prediction accuracy (R\u0026sup2;=0.949, MAE\u0026thinsp;=\u0026thinsp;0.022 for the test set). The RF model was used as the fitness function and a global search was performed using GA to obtain the optimal preparation conditions (5.4% loading, Ca:Mn\u0026thinsp;=\u0026thinsp;1:4, calcination temperature 423\u0026deg;C, and time 3.75h), which resulted in a 93.7% xylene removal rate, an improvement of 2.5% compared to the traditional experimental optimization. Combined RF model and SHAP value analysis revealed that calcination temperature (43.9%) and load (21.96%) were the key control parameters. Experimental results show that GA can avoid falling into local optimum, enhance overall catalyst performance, and reduce carbon consumption. This study provides a new paradigm for the green and low-carbon development of catalysts in the field of heterogeneous catalysis, and mechanisms such as multi-objective optimization can be further explored in the future to promote industrial chain-wide innovation.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eDeclaration of Competing Interest\u003c/h2\u003e\u003cp\u003eThe authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Changming Du reports financial support was provided by National Natural Science Foundation of China. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThe work is supported by the National Natural Science Foundation of China [61871409].\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eA.J. wrote the main manuscript text.W.W. was responsible for visualization and verification.W.Z.Y. was in charge of part of the image drawing.Q.Q.H. proposed the concept.D.C.M. was responsible for the review.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eDu, C. M., Lu, S. Y., \u0026amp; Ding, J. M. (2023). Heterogeneous catalytic oxidation treatment of organic waste gas and engineering applications. Beijing, China: Chemical Industry Press.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGuo, Y. L., Wen, M. C., Li, G. Y., \u0026amp; An, T. C. (2021). Recent advances in VOC elimination by catalytic oxidation technology onto various nanoparticles catalysts: A critical review. Applied Catalysis B: Environmental, 281(2), 119447.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKonsolakis, M. (2015). Recent advances in catalyst technology for volatile organic compound (VOC) removal. Catalysis Today, 258, 13\u0026ndash;23. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cattod.2015.03.041\u003c/span\u003e\u003cspan address=\"10.1016/j.cattod.2015.03.041\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, H., Zhang, L., Li, X., et al. (2022). Non-noble metal catalysts for catalytic oxidation of volatile organic compounds: A review. Applied Catalysis B: Environmental, 310, 121336\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e.https://doi.org/10.1016/j.apcatb.2022.121336\u003c/span\u003e\u003cspan address=\".10.1016/j.apcatb.2022.121336\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZeng, Y. Q., Haw, K. G., Wang, Z. G., Wang, Y. A., Zhang, S. L., \u0026amp; Hongmanorom, P. (2021). Double redox process to synthesize CuO-CeO₂ catalysts with strong Cu-Ce interaction for efficient toluene oxidation. Journal of Hazardous Materials, 404, 124088.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eChen, Y., Li, R., Suo, H., \u0026amp; Zhang, Z. (2021). Evaluation of a data-driven machine learning approach for identifying potential candidates for environmental catalysts: From database development to prediction. ACS ES\u0026amp;T Engineering, 1(8), 1246\u0026ndash;1257.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShih, S., Lee, W.J., Lin, L.F., Huang, C.H. (2012). Prediction of catalytic activity of Cu-Ce/γ-Al2O3catalysts by artificial neural networks. Chemical Engineering Journal, 181\u0026ndash;182, 130\u0026ndash;136. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cej.2011.11.040\u003c/span\u003e\u003cspan address=\"10.1016/j.cej.2011.11.040\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLundberg, S. M., \u0026amp; Lee, S. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30, 4768\u0026ndash;4777.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu, Y., Dai, H., Deng, J., Xie, S., Yang, H., Jiang, Y. (2015). Surface reconstruction of Pt nanoparticles for enhanced CO oxidation: A combined experimental and DFT study. ACS Catalysis, 5(2), 909\u0026ndash;919. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/cs501668k\u003c/span\u003e\u003cspan address=\"10.1021/cs501668k\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLiu, S., Wen, C., Liu, Q., et al. (2024). Theoretical optimization of bed packing arrangement in cascade Dual-Catalyst system with side reactions. Chemical Engineering Science, 299, 120500. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ces.2024.120500\u003c/span\u003e\u003cspan address=\"10.1016/j.ces.2024.120500\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, Z. L., Zhang, H. K., Ren, J. H., \u0026amp; Li, X. (2021). Predicting adsorption ability of adsorbents at arbitrary sites for pollutants using deep transfer learning. npj Computational Materials, 7(1), 19.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJablonka, K. M., Ongari, D., Moosavi, S. M., \u0026amp; Smit, B. (2020).Big-data science in porous materials: Materials genomics and machine learning.Chemical Reviews, 120(16), 8066\u0026ndash;8129.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1021/acs.chemrev.0c00004\u003c/span\u003e\u003cspan address=\"10.1021/acs.chemrev.0c00004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJiao, F., Li, J., Pan, X., Xiao, J., Li, H. (2016). Selective conversion of syngas to light olefins. Science, 351(6277), 1065\u0026ndash;1068. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1126/science.aaf1835\u003c/span\u003e\u003cspan address=\"10.1126/science.aaf1835\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBreiman, L. (2001). Random forests. Machine Learning, 45(1), 5\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi, X., Zhang, Y., Cheng, J., Liu, Y. (2019). Machine learning-assisted design of porous materials for VOC adsorption. Microporous and Mesoporous Materials, 278, 379\u0026ndash;386. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.micromeso.2019.01.010\u003c/span\u003e\u003cspan address=\"10.1016/j.micromeso.2019.01.010\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHastie, T., Tibshirani, R., \u0026amp; Friedman, J. (2009). The elements of statistical learning: Data mining, inference, and prediction (2nd ed.). Springer.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDong, Y. T. (2023). Comparison of Random Forest and Traditional Classical Method in Regression and Classification Problems. Statistics and Applications, 12(2), 255\u0026ndash;260. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.12677/SA.2023.122026\u003c/span\u003e\u003cspan address=\"10.12677/SA.2023.122026\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDrucker, H., Burges, C. J. C., Kaufman, L., Smola, A., \u0026amp; Vapnik, V. (1997). Support vector regression machines. Advances in Neural Information Processing Systems, 9, 155\u0026ndash;161.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMontgomery, D. C., Peck, E. A., \u0026amp; Vining, G. G. (2021). Introduction to linear regression analysis (6th ed.). Wiley.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eShin, Y., Kim, Z., Yu, J., \u0026amp; Park, S. (2019). Development of NOx reduction system utilizing artificial neural network and genetic algorithm. Journal of Cleaner Production, 232, 1418\u0026ndash;1429.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoldberg, D. E. (1989). Genetic Algorithms in Search, Optimization and Machine Learning. Addison-Wesley.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMitchell, M. (1996). An Introduction to Genetic Algorithms. MIT Press.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRezaei, E., Soltan, J., Chen, N. (2013). Catalytic oxidation of toluene by ozone over alumina supported manganese oxides: Effect of catalyst loading. Applied Catalysis B: Environmental, 136\u0026ndash;137, 239\u0026ndash;247. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.apcatb.2013.01.060\u003c/span\u003e\u003cspan address=\"10.1016/j.apcatb.2013.01.060\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWang, Y., Arandiyan, H., Scott, J., Amal, R., Dai, H., Deng, J., Li, J. (2017). Effect of calcination temperature on the activity and structure of MnOx/γ-Al2O3 catalysts for formaldehyde oxidation. Catalysis Today, 281, 467\u0026ndash;472. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cattod.2016.05.044\u003c/span\u003e\u003cspan address=\"10.1016/j.cattod.2016.05.044\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eZhang, L., He, C., Li, Y., Cheng, Y. (2020). Recent advances in catalytic ozonation for water treatment. Chemical Engineering Journal, 397, 125421. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cej.2020.125421\u003c/span\u003e\u003cspan address=\"10.1016/j.cej.2020.125421\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLi, F., Wang, S., Li, T., Tian, Y., Wang, M., \u0026amp; Cai, W. (2023). Effect of Calcination Temperature on the Performance of SiO\u003csub\u003e2\u003c/sub\u003e@Co@CeO\u003csub\u003e2\u003c/sub\u003eCatalyst in CO\u003csub\u003e2\u003c/sub\u003eReforming With Ethanol. Catalysis Letters, 153(12), 3712\u0026ndash;3723. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10562-023-04282-6\u003c/span\u003e\u003cspan address=\"10.1007/s10562-023-04282-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDraper, N. R. Smith, H. Applied Regression Analysis, 3rd Edition[M]. New York: Springer. 2014.\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eJames, G., Witten, D., Hastie, T., \u0026amp; Tibshirani, R. (2021).An Introduction to Statistical Learning: With Applications in R (2nd ed.).Springer.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/978-1-0716-1418-1\u003c/span\u003e\u003cspan address=\"10.1007/978-1-0716-1418-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLundberg, S. M., Erion, G., Chen, H., DeGrave, A., Prutkin, J. M., Nair, B., \u0026hellip; Lee, S. I. (2020).From local explanations to global understanding with explainable AI for trees.Nature Machine Intelligence, 2(1), 56\u0026ndash;67.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1038/s42256-019-0138-9\u003c/span\u003e\u003cspan address=\"10.1038/s42256-019-0138-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eRaccuglia, P., Elbert, K. C., Adler, P. D., Falk, C., Wenny, M. B., Mollo, A., \u0026hellip; Brinson,L. C. (2016).Machine-learning-assisted materials discovery using failed experiments.Nature,533(7601), 73\u0026ndash;76.https://doi.org/10.1038/nature17439\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Catalyst optimization, Machine learning, Genetic algorithms, Multiphase catalytic technology","lastPublishedDoi":"10.21203/rs.3.rs-7555785/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7555785/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMultiphase catalytic technology is an effective method for treating exhaust gases, but the traditional method for optimizing catalyst preparation conditions is inefficient and struggles to achieve multi-dimensional optimization. Aiming at the problems of low efficiency of experimental optimization and tend to fall into local optimum in the development of multiphase catalytic oxidation catalysts, we propose an intelligent optimization strategy integrating machine learning and genetic algorithm (GA). Taking Mn-Ca/γ-Al2O₃ as the research object, the effects of catalyst loading, loading ratio, calcination temperature and time on the degradation of xylene were systematically investigated, and a high-quality dataset comprising 225 sets of experimental data was established. Algorithms such as Random Forest (RF) and Gradient Boosting Tree (GBR) were used to establish a removal rate prediction model, which combined with SHAP values to resolve the key influencing factors, and we used GA to search for the optimal preparation parameters. The results showed that the RF model had the best prediction accuracy (R²=0.949, MAE=0.022 for the test set) and the characteristic importance showed that calcination temperature (43.9%) and load (21.96%) were the key control parameters. The best conditions obtained from GA optimization (5.4% loading, Ca: Mn=1:4, calcination temperature 423°C, time 3.75h) resulted in 93.7% xylene removal, which is a 2.5% enhancement over the conventional experimental optimization. This study provides a new paradigm for green and low-carbon development of catalysts.\u003c/p\u003e","manuscriptTitle":"A method for optimizing catalyst preparation conditions based on machine learning and genetic algorithm","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-27 15:18:47","doi":"10.21203/rs.3.rs-7555785/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b779a53d-121d-4813-9434-92b399f72d89","owner":[],"postedDate":"October 27th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-11-01T06:23:44+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-27 15:18:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7555785","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7555785","identity":"rs-7555785","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}